Battery SOC Estimation Using extended Kalman Filter and ANN with Measurement of Noise

Abstract

Estimation of State of charge (SOC) has always been the crucial segment of battery management system (BMS). However, SOC cannot be determined accurately, it can only be estimated. In this work, an electrical equivalent circuit of the Li-Ion polymer battery of rating 11.5 Ah based on Randel's model has been simulated in MATLAB using the results obtained from a practical model. The non-linear nature of the battery is considered with the model dependent on State of Charge (SOC) and temperature variations. In this paper, Extended Kalman Filter (EKF) algorithm and an Artificial Neural Network (ANN) algorithm have been employed for SOC estimation. Through EKF method, the simulation results show that the error is less than one percent between true SOC and estimated SOC. Through ANN algorithm, when properly trained with sufficient training data - Neural Network consisting of a single layer is capable of adequately appropriating the non-linear characteristics of a battery. It is found that at 271 epochs, it is able to achieve an error as low as 0.010291 where SOC values are varied from 0 to 1. These methods are equally applicable to other battery models.